Interrelationship between Poverty, growth and
inequality in India : A spatial approach
Sandip Sarkara and Samarjit Dasb
a,b Economic
Research Unit, Indian Statistical Institute.
203, B. T. Road, Kolkata 700108, India.
Date of Creation : 17/05/2014
Abstract
The responsiveness of poverty as a result of growth and inequality
has been studied using a panel data of rural and urban state regions
of India, for five consecutive National sample survey round of India.
These regions are combinations of different districts of a state of India. The micro level analysis not only increases efficiency by mere
increment in the number of observations, but also captures intrastate
heterogeneity of growth and redistribution effects on poverty reduction. Since, the state regions used for this analysis are based on fixed
boundaries where there is frequent movement of citizens, we have also
controlled different kinds of spatial dependencies in the model. We
find strong evidence of positive spatial dependence of one region over
the other in terms of poverty reduction. Estimated growth elasticities
of poverty lies mostly in the range of -1.5 to -2.5. and the Gini elasticity of poverty lies in the range of 0.5 to 1.7. The absolute values
of these elasticities are greater for the rural India, compared to urban
India and also increases over time.
Key words: Poverty, Growth, Inequality, Spatial Approach, India
1
1
Introduction
India is one of the largest growing economy in the world. During the last
two decades, she has not only been able to maintain a sustained growth, but
also reducing poverty steadily. However, neither growth rate nor poverty
reduction, is uniform across regions of India. The non-uniformity might be
either due to economic growth, or due to different aspects of poverty-reducing
impact of that growth.
Our objective in this article is not only to explore the role of growth, but
also on the role of distributional effects of income distribution, on poverty reduction. Thus we will also focus on the fact, whether the reduction of poverty
is embedded due to unequal incomes. Addressing this problem is not new
in the literature, and many theoretical and empirical researches have been
done in this direction. The central theme of research agenda in this context is based on estimation of a summary index called growth elasticity of
poverty (GEP). The indicator GEP is important in terms of policy prescription in the sense that it captures the responsiveness of poverty as a result of
increase (decrease) of 1% growth. Datt and Ravallion (1992) suggested an
approach based on the decomposition of poverty reduction on growth and
inequality components. For application of the methodology it is necessary to
have information on the whole income distribution. In the context of cross
country studies, usually such data sets are unavailable in most of the cases.
Ravallion (1995) considered a model with poverty reduction was considered
to be a function of growth rate of average income. Further the model was
generalized by considering the fact that income is endogenous in the poverty
estimating equation. However, the model is based on cross sectional observations, fails to capture the country specific effects. These effects are likely to
be correlated with growth rate, which would lead to biased and inconsistent
2
estimates of the parameters. In order to overcome this problem it is necessary to consider a model based on panel data. There are studies based on
panel data for estimation of GEP(Ravallion and Chen, 1997; Adams, 2004;
Ram, 2007; Chambers and Dhongde, 2011). It has been observed that in
most of the studies, where poverty reduction is regressed on inequality and
growth, the value GEP lies in the range -2 to -4 and IEP as positive.1
However, all these studies are based on a costlier assumption that the
growth elasticity of poverty to be constant. Even if the model includes both
mean income and the Gini index as linear regressors, it does not interact these
explanatory variables, which effectively prevents inequality from affecting the
magnitude of the estimated GEP. Assuming log normality of income distributions Bourguignon (2003) develops an econometric model for the poverty
estimation equation. He begins with an econometric model where growth rate
of poverty is regressed on growth rate of inequality and income, along with
the interactions of both these variables with initial income inequality and the
ratio of poverty line and mean income. The later variable was refereed to be
inverse development factor. Like GEP another indicator is also important
which captures the responsiveness of inequality on poverty, popularly known
as inequality elasticity of poverty (IEP). Fosu (2009) also considered this
model. He has explored the extent to which inequality influences the impact
of growth on poverty reduction, based on a global sample of 1977-2004 unbalanced panel data for Sub Sahara African (SSA) and non-SSA countries.
1
Ravallion and Chen (1997) and Adams (2004), considered not only growth rate of
income but also have included growth rate of gini (as a proxy of a relative inequality
measure) as explanatory variables. Adam’s have shown that GEP differs substantially
when growth rate of mean income and GDP, is considered as a measure of growth rate.
Chambers and Dhongde (2011) considered a model which considers the nonlinearity of
growth-poverty-inequality nexus, by considering a non-parametric regression model. Ram
(2007) model is different in the sense that instead of considering the poverty estimation
equation in growth terms, all the variables are based on their level values. The estimated
GEP and IEP for this model is substantially different from the others.
3
It was also shown that impact of GDP growth on poverty reduction is a decreasing function of initial inequality. The study additionally observes that
higher rates of increases in inequality tend to exacerbate poverty, with the
magnitude of this effect rising with initial income. Kalwij and Verschoor
(2007), generalized the model with further specifying income as endogenous
variable. The endogenity was considered mainly because of the fact that income growth rate and poverty indexes are computed from the same variable.
Although, Kalwij considered a generalized model, however, the sign and significance of the estimates is similar to that of Bourguignon (2003) and Fosu
(2009). Al the studies discussed above are based on cross country data, and
thus criticized following the weak comparability of primary survey rounds in
most of the cases, for details see Ravallion and Datt (2002).2
The literature discussed above, clearly shows academicians have given
immense importance to the estimation of these elasticities. A combination
of GEP and IEP, might be very helpful for policy makers, in the sense that
it helps to understand the poverty responsiveness respectively due to growth
and redistribution. The objective of this article is to evaluate the “povertygrowth-inequality” nexus of India, using these elasticities. However, the
major departure from the literature, is this study is based on a micro level
panel dataset, consisting of the rural and urban regions of different states
of India. These regions are the smallest possible stratum considering the
multi-stage National Sample Survey Office Data (NSSO) sampling schemes.
Clearly comparability is not an issue in this regard since the units we consider
are independent stratum and the survey design has remained unchanged in
this period. No study (as per our knowledge), has focused on such a micro
2
GEP has also been estimated for country specific data e.g Ravallion (2011) had estimated the GEP separately for Brazil, India and China. Zaman and Khilji (2013) has
estimated the GEP for Pakistan.
4
level analysis on this topic. An alternative approach would have been consideration of state level analysis following (Ravallion and Datt, 1998, 2002).
It should be noted that this micro level analysis is not just merely increasing
the efficiency, by the increment in the number of observations. It has been
observed that in India’s one of the largest state Uttar Pradesh, the percentage of poor in the western part is 34%, which on the eastern part is much
higher. The hetrogenity within the state will be missed, if we consider a
state level analysis. Further we will also see as we move through this article,
that within state, estimates of the GEP and IEP also varies substantially.
The heterogeneity of poverty rates of Uttar Pradesh, might be due to spatial dependencies of the units of analysis. Since, the western part shares a
common boundary with Delhi, the development schemes of country’s capital
might have been trickled down to its neighbour. There are many other examples on this direction, which further motivates us to consider an econometric
model with spatial dependencies. Ignoring these dependency, would lead to
biased and inconsistent estimates of the parameters, See Anselin (2009).
Inclusion of the spatial dependence also makes our model special in this
context. As per our knowledge, no study considered these dependencies so far
on estimation of GEP and IEP. We will begin with a Bourguigon type model.
However, we will also include policy variables like education and indoor air
pollution in the poverty estimation equation. We will further consider income
to be an endogenous variable and could generalize the model substantially.
Although this exercise is similar to Kalwij and Verschoor (2007), however
since their study is based on cross country data, the set of instruments are
different for the two analysis.
The paper has been included in the following fashion. In section 2 we
provide a brief description of a general Bourguigon type model and related
5
issues. Section 3 provides a brief description of data and also on computation of poverty rates and inequality measures. In section 4 we discuss on
incorporation of spatial dependencies. Section 5 we discuss briefly on econometric models. A general model with further considering the problems of
endogeneity has been discussed in section 6. In GEP and IEP computation
details along with interpretation of the results.
2
Econometric Model
We will use a Bourguigon type model (Bourguignon, 2003). Assuming income
follows a log normal distribution, the elasticities are computed analytically.
The analytical results were further extended by Kalwij and Verschoor (2007),
it was seen that the growth and inequality elasticity of poverty, (which we
call GEP and IEP), depends on the initial inequality and ratio of poverty line
and mean. The last factor is also commonly known as inverse development
factor.
Let Pit , Yit and Git , represents the poverty, average income and income
inequality of region i ∈ {1, 2, ...N }, at time point t ∈ {1, 2, ...T }, and pit ,
yit , and git as their growth rate3 . Poverty estimating equation following
Bourguigon in a panel data context may be written as follows4
pit = θi +α1 yit +α2 yit i0 +α3 yit Yit /Z +β1 git +β2 git i0 +β3 git Git /Z +uit (1)
3
For any variable X, the denote the growth rates as x, or x = ∆log(X)
Bourguignon (2003) started with a naive models as pit = β0 + β1 yit + uit . A Standard
Model, was also proposed may be written as follows pit = β0 + β1 yit + β2 git + uit . It
was noticed that R square, increases as one moves from the naive model to the standard
model. R square is almost doubled if one moves from the standard model to the model
specified in 1.
4
6
where Z and Gi0 are the poverty line and initial income inequality(gini
coefficient). θi is the unobserved panel heterogeneity.5 Consider X as the set
of explanatory variables as denoted in equation 1. In matrix notation we can
write the model as follows
p = Xβ + u
(2)
where X is the set of all exogenous variables and u is the residual, with
usual OLS assumptions. From now we will refer X as the set of Bourguigon
variables. The rationale for incorporation of this model is twofold. Firstly
non linearities of the relationship between poverty-inequality-growth to some
extent.6 Secondly, we will show in the next section that GEP and IEP are not
fixed and depends on the initial inequality and
Z
Y
ratio. Thus it is possible
to capture the heterogeneity of the growth-poverty-inequality relationship
across regions.
2.1
GEP and IEP : Functional forms
GEP and IEP are the responsiveness of poverty reduction, respectively for
increment of 1% growth rate and income inequality. Once αi and βi ∀i ∈
{1, 2, 3} are estimated from 1, GEP and IEP turns out to be
5
On further assumptions on whether θi is correlated with the explanatory variables or
not, i.e. whether a fixed or a random effect model is considered, we will discuss latter.
6
A better way to capture the non-linearities of the relationship is to adopt a non
parametric estimation equation, similar to Chambers and Dhongde (2011). However, since
the state regions are based on fixed boundaries it is likely to reflect spatial dependencies
among each other. Ignoring the spatial dependencies (if exist) would lead to biased and
inconsistent estimates of the parameters. Inclusion of a non parametric model along with
the spatial effects is beyond the scope of this article.
7
GEP = α0 + α1 i0 + α2 (Z/Y )
(3)
IEP = β0 + β1 i0 + β2 (Z/Y )
(4)
The values of GEP and IEP depends on the combination of initial inequality and inverse development factor (Z/Y ratio). It should be noted,
that so far or discussion, has been limited that the explanatory variables are
uncorrelated with the residuals. However, presence of income growth rate
(y), really questions this assumption. Such endogenity would lead to biased
and inconsistent estimates of the parameters and consequently for GEP and
IEP. We will come to this issue latter.
2.2
Decomposition methodologies and the econometric models :
Datt and Ravallion (1992) suggested a methodology on the decomposition of
poverty reduction into three components, viz. growth, redistribution and the
residuals. The growth (redistribution) component is defined as the change
in poverty due to change in mean income (Lorenz Curve) holding the Lorenz
curve (mean income) constant. The residual is the difference between the
growth (redistribution) components evaluated at the terminal and initial
Lorenz curves (mean incomes), respectively. lf the mean income and/or the
Lorenz curve remains unchanged over the decomposition period, then the
residual vanishes. These approaches have been considered to be more ideal
particularly in cases where data on whole distribution is available (Bourguignon, 2003; Kalwij and Verschoor, 2007; Chambers and Dhongde, 2011).
A recent decomposition approach, based on Shapley value has been suggested
8
by Shorrocks (2013).
In spite of the fact that we have informations on the whole distribution,
we will not use the decomposition analysis in this article. Firstly, in these
methodologies it is not possible to account the responsiveness of different
policy variables like education and health. Similarly the state specific effects
are also not accounted in the decomposition methodologies. In a panel data
set up it is possible to control these effects. In the context of decomposition
methodologies both these effects are likely to be accounted in the residuals.
The results of the decomposition are questionable when the residual component is not small relative to the change in poverty7 . The micro data set
that has been considered for these analysis, is likely to reflect spatial dependencies. For example, when poverty of a region declines people migrates
there, and thus it is expected that poverty of both the regions is likely to be
affected.8 In the decomposition methodologies for two distinct regions i and
j, poverty reduction of i and j are independent and consequently ignores the
spatial dependencies.
However, it is not true that the econometric model discussed in 1, is
free from any criticism. An alternative functional form and inclusion of
some additional possible variable, might actually change the GEP and IEP
estimates substantially. It is also possible that growth and poverty both being
estimated from the same variable income, any measurement error of income
is likely to effect poverty rates also. Many policy variables might affect the
poverty reduction rate. For those policy variables, data is unavailable, would
be accounted in the residuals. It is likely that those variables might also be
correlated with the growth rate. Thus it seems it is necessary to consider
the growth rate as an endogenous variable and model accordingly. We will
7
8
See Growth-Inequality Decomposition, in the PovertyNet website of World bank.
We will discuss in details on the issue of migration and poverty latter.
9
discuss a generalized model with two endogenous variable, viz. spatial lag of
the poverty rates and growth rate.9
3
Data
We have created a balanced panel data set from the five consecutive National
Sample Survey Office (NSSO) quinquennial rounds. The quinquennial rounds
also known as the thick survey rounds usually conducted after every five years
for the estimation of poverty rates and employment-unemployment status of
India. In this analysis we consider 43rd, 50th, 55th, 61st and 66th round
data which provides data for the period of July 1987 - June 1988, July 1993
- June 1994, July 1999- June 2000, July 2004-June 2005 and July 2009-June
2010 respectively. The panel variable for this analysis is a NSSO stratum
called state regions, which are combinations of different districts of a state.10
However, it should be noted that the number of districts and also states has
changed over time and consequently NSSO definition over the period has
also changed. In order to maintain the regional identity over the period we
have to merge more than one state regions in many cases.11 The number of
modified state regions are 128, of them 64 are rural state regions and the rest
are urban state regions.
The main variable needed for establishing the empirical relationship be9
Like the income it is also possible that growth rate of relative inequality is also
likely to affect the poverty rate, however, it is difficult to estimate a model with so many
endogenous variables.
10
It seems that a possible option would have been considering an analysis at a district
level. However, in the multi-stage sampling design districts are not stratum. It is possible
that a given district may not have adequate observations. Further the representations of
all sections of the society is also not guaranteed. Combing these two factors we have every
possibilities of getting a biased and inconsistent estimate.
11
If an estimate is consistent for two independent stratum, the estimate is also consistent
even if we merge the two independent stratum. Hence merging the state regions wont
create a problem, from the point of sampling design and other related issues.
10
tween growth poverty and inequality is income. However, data on income is
not available in India, thus expenditure is considered to be a proxy. From
now on by income we mean monthly per capita expenditure or MPCE. NSSO
conducts a program of quinquennial surveys on consumer expenditure provides a time series of household consumer expenditure data, which is the
prime source of statistical indicators of level of living, social consumption
and well-being, and the inequalities thereof. MPCE is obtained from the
NSSO quinquennial surveys on consumer expenditure. In order to collect
data on Monthly per-capita expenditure, one must set a recall period of
consumption. We will use monthly consumer expenditure on the basis of
a mixed recall period data.12 As we have mentioned earlier at some stage
of this article we will focus on issue of endogeneity of the income growth
variable. Most of the instruments for this analysis are obtained from NSSO
employment-unemployment rounds. The employment unemployment rounds
are also conducted in the same period and similar survey design. Thus it is
possible to obtain estimates for the same state regions.
3.1
Computation of Poverty growth and inequality
The first exercise for the poverty estimation of a society is the specification
of poverty line. Since, poverty line for the state regions are not available. We
have used state specific poverty line for the state regions. There has been
a change in the methodology of estimating the poverty line, since after the
publication of Tendulkar committee report in 2004-05. In order to maintain
consistency over periods, we have inflated the poverty lines, using Consumer
12
In a mixed recall period, data for educational, medical (institutional), clothing, bedding, footwear and durable goods are collected on a recall period of 365 days. The other
items are collected on the basis of a recall period of 30 days. We have used scheduled type
1 data for 66 th round in order to maintain comparability. For details see NSSO reports.
11
price indices for agricultural labor (CPIAL) and Consumer price indices for
industrial workers respectively for rural and urban India. Real MPCE for
both rural and urban India, are obtained using these price indices. The
growth rate of average real MPCE of rural and Urban state regions are
considered to be the proxy of average growth rate of the society.
3.2
Policy Variables
Education : It is possible that a society is able to combat poverty better
if the number of literates are higher, for details see Gundlach et al. (2004).
Education may increase growth and also have an effect on inequality.13 Thus
ignoring this variable would lead to endogeneity problem in the form of omitted variable bias.
NSSO provides data on literacy status of all the individuals coded in
different groups v.i.j primary, secondary, higher secondary and graduates
and above. We have considered percentages of female adults (aged 15 years
or more), having secondary level of education as a proxy of the education
variable.14 We expect a negative coefficient for the education variable, in the
poverty reduction equation.
Air pollution and health hazards through energy consumption (Indoor air pollution) Air pollution might be broadly classified by two different phenomenon v.i.j outdoor phenomenon and indoor phenomenon. The
outdoor phenomenon is largely due to the smoke produced by the factories,
13
It is logical to think that a society with higher number of literates might lead to higher
growth rate. Presence of large number of literates might lead to higher productivity
and consequently higher growth rates. The chain of education and inequality can also
be derived logically, if we focus on the one to one relationship between corruption and
inequality as pointed out by Sung and Khagram (2005). Educated people might actually
raise voice against ground level corruption which often directly affects the poor.
14
We have considered female literacy rates as a proxy of education following Ravallion
Ravallion and Datt (2002).
12
mostly takes place in the industrial areas. In developing countries this is often classified as an urban problem. In rural areas people use bulk of the fuels
burned (by mass) are solids, principally wood and coal. Unlike gases and liquids, solid fuels require relatively advanced technology to be pre-mixed with
air or otherwise ensure their complete combustion. The airborne emissions
of incomplete combustion products, such as carbon monoxide, particulates,
and volatile organic compounds, have been extensive. For more details see
Smith (1993) and the references cited there in. The list of health hazards as
a result of the indoor pollution, that has been documented by Smith are as
follows
1) Respiratory infections in young children
2) Adverse pregnancy outcomes for women exposed during pregnancy
3) Chronic lung diseases and associated heart disease in adults and
4) Cancer.
Given the data sets it is not possible to capture the outdoor smoke factor. However, NSSO collects data on principle source of cooking, which
might be considered as an indicator of indoor air pollution. We consider the
percentages of people effected directly from indoor air pollution as another
explanatory variable.15 A better indoor environment might increase physical
abilities of individuals and thus help them combating poverty.
3.3
Descriptive Statistics
In Table 1 and 2 we have presented the average values of poverty, income
inequality, average MPCE, % of households having electricity, female literacy
rates, % of cultivated land in the month of June and July, cultivation and
15
The % of individuals using one of the following coke, coal, firewood and chips, dung
cake and charcoal are assumed to be sufferer of indoor air pollution.
13
% of households whose chief source of cooking fuels are prone to causing
different health hazards and also indoor air pollution, respectively for rural
and urban India.
Less developed states like Bihar, Orissa, Madhya Pradesh, Chattisgarh
and Jharkhand, shows poor performance in most of the indicators. The
poverty indices are also high for these states. The similar pattern is also
observed in the highly developed states like Punjab, Haryana, Delhi Chandigarh e.t.c. These states are also contiguous to each other and performs better
in respect to almost indicators. In Southern and western states developed
states are Karnataka, Kerala, Tamil Nadu and Maharastha. They are also
contiguous in most of the cases. This pattern indicates the presence presence
of spatial dependence among the variables, including the poverty indices.
Although states seems to exhibit a spatial pattern, however, intra state
inequalities are also inevitable in many cases. For example the Rural Malwa
regions of Madhya Pradesh shows average HCR 37.36%, whereas in the same
state the south eastern part exhibits a poverty rate of 66.05%. Similarly the
western part of Uttar Pradesh (contiguous to Delhi) exhibits a poverty rate
of 36% part where as the other regions shows much higher poverty rates.
As expected the air pollution factor captured by the usage of cooking
fuels material shows huge difference between the rural and urban regions. In
all cases more than 80% of the individuals use cooking fuels harmful a to
health. The only exception being Delhi and Chandigarh where in the rural
areas this rate is less than 20%. Percentages of households having access to
electricity is better in most of the cases in urban India, but in rural India
performance of some states are extremely poor, e.g Jharkahand, Central
Uttar Pradesh, Southern Orissa e.t.c. The variation of the gini coefficent has
not been noticed much.
14
Insert Table1 and Table2 here
4
Spatial dependencies
The rationale of spatial econometrics is based on Tobler’s first law of geography states that “Everything is related to everything else, but
near things are more related than distant things”. Poverty estimating equation 1, is based on the assumption that all the all observations are
independent and identically distributed (iid). Although the state regions are
based on specific boundaries, but the citizens move from one place to another often, it is possible that they reflect spatial dependencies, where values
observed at one location depends on others. It is also possible that policies
of one state region may also be reflected to another very quickly, would lead
to the spatial dependencies.
In order to capture these dependencies, in terms of an empirical model,
one must specify a spatial weight matrix. We consider a contiguous weight
matrix, which takes values 1, if two region are contiguous (neighbours) to
each other, else 0. Let WN = {wij } be a square matrix of spatial weights of
size N × N , N is the number of regions, where
wij = 1
= 0
(if i and j are contigous and i 6= j)
(else)
(5)
Poverty reduction , might be modified by incorporating spatial dependencies in the dependent variable and/or in the residuals.
Spatial dependence in the dependent variable : A modified version
of equation 2 with spatial dependencies of the dependent variable may be
15
written as follows
p = ρW p + Xβ + u
(6)
The spatial autocorrelation variable is endogenous in the above equation.
Thus, OLS estimation of equation 6 leads to a biased and inconsistent estimation of the parameters. However consistent estimation of the parameters
are possible following a Maximum Likelihood method of estimation(MLE).16
In case ρ is statistically significant, but ignoring the fact we consider a model
without the spatial lag variable, would also give biased and inconsistent estimates.
Spatial dependence in the error terms : It is not always necessary
that the spatial dependencies is reflected only in the dependent variable. We
can also consider a model with spatial dependencies in the residual series
and/or in the dependent variable as follows
p = α + ρW p + Xβ + u + λW2 u
(7)
where W2 , denotes the spatial matrix that captures the spatial dependencies of the residual series.17
Ignoring the spatial dependencies in the residual series would lead to inconsistent estimation of the standard errors. However, unlike the SAR model
even if the spatial dependencies are ignored estimates of the coefficient would
be unbiased and consistent. In some situations it is possible that the residuals are cross sectionally dependent, and/or violates the usual assumption
of OLS (autocorrelation and Heteroskedasticity problems). In order to deal
16
See Anselin (2009) for further details.
Although it is possible to consider a different weight matrix for the dependent variable
and residual series. However, in this case we will consider a simple model with a same
spatial weight matrix.
17
16
with such situations one may also use Driscoll Karry Standard errors
(DSK SE)Driscoll and Kraay (1998).
We will estimate the spatial models both with and without consideration
of the fact that income growth rate (y) is endogenous. In the first case, where
‘y’ is assumed to be exogenous the model will be estimated by the usual
Maximum likelihood estimation methodology suggested by Anselin (2009).
In the second case we have two endogenous variables in the right hand side
v.i.z spatial lag and y, is estimated by General method of moments. The
methodology of GMM in the presence of spatial dependencies was introduced
by Kelejian and Prucha (1999). Kelejian and Prucha (2004) extends the
GMM methodology by including additional endogenous variable (excluding
spatial lag).
5
Econometric Results
In Table 4 we have presented the estimates of the econometric model as specified in Equation 2. For the robustness of the analysis we have considered
three different poverty indices viz. HIC, PG and SPG. It can be readily observable that the set of exogenous variable not only contains the Bourguigon
variables, but also policy variables like cooking and literacy rates. In order to
estimate the standard errors consistently, we have used the SAR models with
Driscoll Karry Standard errors (DSKSE)Driscoll and Kraay (1998). DSKSE
captures all kinds of cross section and temporal correlation, of the residuals.
Another option would have been consideration of spatial dependencies not
only in the dependent variable, but also in the error part as in equation 7.
However, we find insignificant λ in all cases, thus incorporating such models
would lead to inconsistent estimation of the standard errors. It should be
17
noted that the policy variables female literacy and cooking fuels has been
found to be significant in most of the cases. The signs are also appropriate.
It has been noticed that the spatial autocorrelation parameter is positive
and highly significant for all the cases. As we have mentioned earlier, ignoring this dependency would lead to biased and inconsistent estimates of the
parameter. The positivity of the spatial autocorrelation parameter implies
that poverty rate of a region is positively related to its neighbour’s poverty
rate. For example, ρ = 0.17, implies poverty of a region increases by 1.7%
if its neighbours poverty increases by 10%. One possible explanation for the
positivity of ρ may be because of migration. When poverty of a region increases, poor people migrates to the neighbouring region and consequently
poverty of that region also increases. If we observe closely the value of the
parameter declines, as we shift our dependent variable from HCR followed by
PG and SPG. This might be possible due to migration of the richer poor or
individuals whose income being close to the poverty line.18 If a poor people
enters in a society as a migrant, HCR would always increase, ignoring the
distance between migrants income and poverty line. However, since PG and
SPG are distributive sensitive would depend on income of the poor. In the
appendix of the paper some analytical derivations has been done to relate
migration and the growth rate of the three poverty indices.
The sign and significance of the first six variable matches exactly to the
earlier articles based on this model (Bourguignon, 2003; Fosu, 2009; Kalwij
and Verschoor, 2007). It should be noted that the coefficients corresponding
to y(g) is not GEP(IEP). For computation of the elasticities we will require
these estimates. We will come to this part later.
It might be readily observable that the coefficients of the interaction ef18
This is the most common type of migration in most developing economies, for further
details see Du et al. (2005).
18
fects respectively for growth and inequality are of opposite signs. Positivity
of the interaction of the growth rate and initial inequality implies higher
initial inequality and/or ratio of poverty line and mean income, reduces the
responsiveness of growth on poverty reduction. The negativity of the inequality terms are also intuitively justified. Societies with higher initial inequality
and/or poverty line mean ratio, reduces the effects of IEP.19
6
Endogeneity Problems ?
The reported results of Table 4 are based on the assumption that the residuals
are uncorrelated with the explanatory variables. Presence of income growth,
however, questions this assumption. We suspect the endogenity of income
growth rate mainly because of the following reasons.
1) Poverty indices and income growth rate are computed from the same
variable MPCE. Any measurement error of income might also be responsible
for measurement error of poverty index.
2) As Deaton (2003) pointed out the participants of the rich are usually
lower in the survey. This clearly will effect both poverty and growth.
3) Poverty of a society depends on many unobserved components which
is accounted in the residuals. Since, the model is based on monetary poverty,
it is likely that these components not only effects poverty rates, but also
growth rate of income.
In order to deal with the problem it is necessary to find a set of instruments which are uncorrelated with poverty but are highly correlated with
growth rate of income. In order to test the whether the income variable is
19
For more details on these coefficients, See Kalwij and Verschoor (2007). They have
computed the analytical forms of the elasticities are assuming income distribution follows
a log normal distribution.
19
endogenous, we will follow an algorithm discussed in details in Gong et al.
(2005).
Step 1 :
Let z be the set of explanatory variables as in equation 1
excluding the growth rate of income or yit . Also assume that there exists a
set of instruments z ∗ , not belonging to z, which are uncorrelated with the
error term of the equation 1 and highly correlated with yit . Let Π denotes
the stacked matrices z and z ∗ . The first stage regression
y = θi + ν 0 Π + ζ
(8)
where ν is the vector of the parameter and ζ is the residual with usual
OLS assumptions. Let ζ ∗ be the estimates of the residual series.
Step 2 :
In the second step we modify equation 1, with an additional
explanatory variable as ŵ, along with spatial dependencies of the dependent
variable.
p = θ0 z + δy + γζ ∗ + v
(9)
If γ is significant implies y is endogenous in the above equation.20
6.1
Set of Instruments
In order to choose the instruments we consider three different development
indicators of each state regions, which are 1) Employment 2) Infrastructure
and 3) Technological Progress. It is logical to assume that improvement of
these indicators will increase the income growth rate. Further we also assume
that these indicators have no direct role in poverty estimation.
20
For details see Gong et al. (2005) and the references cited there in.
20
6.1.1
Employment
Clearly income and employment are related in an one to one basis. We will
consider two main indicators in order to account the employment facilities
of a state region, which are average wages (and/or salaries) and Jobs. The
job opportunity variable is captures the percentage of employment of individuals, except of being labourer of any kind.21 The second variable Wage
and/or salary might be considered as the best proxy for income, provided
informations have been correctly provided.22 The estimates of percentage of
households having jobs and average salary are obtained from NSSO employment Unemployment round.
6.1.2
Infrastructure
We consider agriculture and electricity as a proxy of the infrastructure variable. We do not have data on agriculture at the state-region level, and thus
we consider %ge of land that has been cultivated as the proxy. The second
proxy is electricity in the form of % of households having access to electricity.23 We have obtained data on both these variables from NSSO consumer
expenditures schedule.
21
In the employment unemployment surveys, data on employment status of all the
members of a household is available. National Classification of occupation was provided
by government of India, in 1968. These codes are also known as NCO codes. NSSO collects
data on NCO codes for all the members of household. Using NCO codes for the last seven
days, the jobs variable has been created. It captures the percentage of individuals at the
age 18-60, either not related to any elementary occupations, workers or are labourers.
22
Data are collected for the wage and salary received either in cash or in kind (in terms
of cash), for the last seven days. However, since, self employed individuals does not receive
any salary or wage, data are missing for them. Average wage/salary for the individuals has
been computed excluding self employed individuals. The main problem with this variable
is misreporting of actual wages/salaries.
23
The proxy that has been used might be considered as an crude indicator. It would
have been better if it is possible to incorporate the number of roads schools, colleges or
area of highways and other developed roads. However, it is difficult to obtain data in the
state region level.
21
6.1.3
Technological Progress
The questionnaire on the household level survey conducted by NSSO consumer expenditure and employment-unemployment has no direct question
that might be related with technological progress. We consider % of child
labour as a proxy of technological progress, as obtained from employment
status of children following the NSSO employment-unemployment rounds.
As pointed out by Hazan and Berdugo (2002) in the early stages of development, the economy is in a development trap where child labour is abundant,
fertility is high and output per capita is low. As a result of technological
progress, wage differential between parents and child increases consequently
leads to the decline of child labour and would also lead to an increment of
child education. For all the state regions, it has been observed that, child
labour has been declining over time. State regions with faster decline of child
labour might be due to higher technological advancements.
6.2
Endogenity tests : Results
In Table 5, we have reported the endogeneity tests result for the choice of
three poverty indexes discussed earlier. It might be noticed that the coefficient for the residual (ζ ∗ ) is highly significant for HCR and PG (respectively
at 1% and 5% level of significance). However, for SPG the coefficient turns
out to be insignificant.
Given that, we can not reject the fact income growth is exogenous in
equation 2 (except for SPG), estimates in Table 4 should be reported with
caution. In Table 6 we have estimated the model considering income growth
as endogenous, however, results remain more or less same.
22
7
Growth and Inequality Elasticity of Poverty
In this section our main target is computation of GEP and IEP, following
equation 3 and 4. In order to compute these elasticities we will use αi and
βi ∀i ∈ {1, 2, 3} from Table 6, where growth rate of income is considered as
an endogenous variable.24
GEP and IEP are not constant, these elasticities depends on the initial
inequality and on the ratio of poverty line and mean income ( YZ ). Thus
it is necessary to specify values of initial gini and the
Z
.
Y
We compute the
elasticities for each state region and time points where the initial gini is based
on the gini of 43 rd round, and
Z
Y
for the respective time points. In Table
7, we have reported the average values of GEP and IEP for different time
points, where the average has been taken over rural and urban state regions.
As we have mentioned at the beginning of this article the expected sign
of GEP is negative, implying as a result of growth, poverty declines. The
estimated elasticities in the cross country studies mostly lies in the range -2
to -5 (Adams, 2004). Although this bound is failed to achieve in most of the
cases, specially in rural India, however, the absolute GEP are grater than 1.
This clearly indicates in each period of these survey rounds, 1% increment
in income growth leads to greater reduction of poverty.
IEP is positive (except for a very few cases), in fact IEP close to absolute
values of GEP in many cases. The positivity of IEP implies the adverse
effects of inequality somehow reduces the force of growth to reduce poverty.
One such adverse effect of inequality is the inter relationship between income
inequality and corruption.25
24
Since, the estimated results of the models with or without endogeneity of income
growth rate is more or less same, the GEP and IEP does not changes much for considering
the estimated parameters from the model with out endogeneity.
25
Sung and Khagram (2005) has shown that corruption is related to greater inequalities,
23
It might be readily observed that the absolute values of both GEP and
IEP increases with time. This is possible because in this period mean income
has increased substantially leading to decline of Z/Y ratio. It has also been
observed that the absolute value of this elasticities increases as we move from
HCR to PG and to SPG. This result are intuitively justified if we focus on
the axiomatic literature of poverty measurement. Head count ratio is a naive
indicator and thus gives equal weight to all the poor’s. Thus even if income
of an individual increases, HCR may remain constant if she is unable to cross
the poverty line. Both PG and SPG would decline as a result of such changes
in income distribution. This property is also widely known as Monotonicity
axiom as suggested by Sen’s seminal article (Sen, 1976). SPG is more general
in this regard, since, it responds to transfer of income among the poor, widely
known as the transfer axiom.
In Figure 1, we have plotted the average of GEP and IEP over the time
points, for all the regions. For the ease of interpretation we have taken
the absolute values of GEP. Thus the black portion of the bar bar, shows
the difference between GEP and IEP. It might be readily observed that the
length of urban bar is greater then that of the rural, in almost all the cases.
This implies absolute value of GEP is grater for the urban sector. However,
IEP is also high for the urban areas, in fact the gap between the elasticities,
reflected in the black portion of the graph, is higher for the rural areas in
and the adverse effect is larger in democratic countries. Corruption on the other hand
might directly effect on many policies against the poor. Government of India considers a
programme of targeting the most needy, a measure was developed by which families were
categorized as living “Below the poverty line”. Identified rural families that are below
the poverty line are eligible for government support such as subsidized food or electricity
and schemes to construct housing and encourage self-employment activities.26 As pointed
by Hirway (2003), “the rich and powerful in a village frequently pressurizes the talati and
the sarpanch to include their names in BPL lists”. Thus in a society with higher income
inequality, instead of the poor households, rich households receive the benefits. In some
cases the talaits also claims money which the poor are unable to pay.
24
most of the cases. This naturally implies, the force of poverty reduction in
urban India, as a result of higher growth rate is largely embedded due to
presence of unequal incomes.
Although, low IEP is desirable, in some cases economies with very low
average income leads to negative IEP, for details see Kalwij and Verschoor
(2007).27 For example in Rural India, we find evidence of negative IEP in
three cases v.i.z southern regions of Orissa, South western regions of Madhya
Pradesh and in hilly areas Manipur. The values of IEP for these regions are
respectively -0.82, -0.20 and -0.07. Out of these three regions the southern
part of Orissa is famous for the famine of Kalahandi, which took place in
the 1980s, in the districts of Kalahandi. This region historically suffers from
low growth rate particularly because of the deterioration of the agricultural
conditions.28 South Western Madhya Pradesh, is also a drought prone region
in this neighbourhood.
8
Conclusion
In this paper we have studied on the responsiveness of poverty as a result
of growth and inequality, following a Bourguignon (2003) type model.The
27
Kalwij’s formulation was based on the analytical form of the IEP. The negative IEP
may be logically derived following a hypothetical example. For the sake of simplicity,
imagine a situation that growth rate of a society is zero. Further in that situation imagine
as a result of redistribution of incomes such that inequality increases but poverty decreases,
leading towards negative IEP. For example as a transfer of income from the poorest poor
to a richer poor, such that the later cross the poverty line. In this situation inequality
increases and as consequence poverty declines.
28
Historically these areas are known as drought prone, with low rainfall over decades.
Low agricultural production in this region also has lead to different types of aids and
supports from the government in terms of food aid. This however, lead to further decline
in agricultural production incentives and also agricultural prices. Since, rural India is
mostly related to agricultural productions, income growth rates also behave accordingly
with the deterioration. For further details on the history of the Kalahandi famine, See
Pradhan (1993).
25
study is based on a balanced panel data set from five consecutive National
Sample Survey Office(NSSO) data for the quinquennial rounds. The panel
variable used for the study rural and urban NSSO state regions, which are
lowest possible NSSO stratum. Unlike most of the studies based on cross
country data, comparability of data in terms of survey design and also other
related is not an issue in this context. However, the state regions are based
on specific boundaries and movement of citizens are frequent, thus it is likely
that poverty of a region is spatially dependent to her neighbor. We have
generalized the model by incorporating different spatial dependencies. In the
poverty estimating equation we have also incorporated policy variables like
female literacy rate and indoor air pollution via cooking. For the robustness
of the analysis, we consider three poverty indexes belonging to the class of
FGT(Foster et al., 1984) family v.i.z Head Count Ratio (HCR), Poverty Gap
(PG) and Squared Poverty Gap (SPG). Two alternative models separately for
growth rate of income being exogenous and endogenous have been considered.
The findings might be summarized as follows
1. Our main estimation results are based on the spatial autoregressive
models. It has been found that the spatial autocorrelation parameter
ρ is positive and highly significant for any choice of poverty indexes
belonging to the FGT class. Ignoring the dependencies would had lead
to biased and inconsistent estimates of the parameter. The results remains unchanged even if we consider income growth as an endogenous
variable. The possible reasons for the positivity of ρ is migration. It has
been observed that ρ declines as we move from poverty indexes being
more sensitive to the income of poor. Du et al. (2005) has pointed out
that among the poor, poorest of the poor often fails to migrate. The
decline of ρ from HCR −→ P G −→ SP G might also be explained by
26
the migration of the persons close to the poverty line. The last statement has been considered as a proposition and we have shown growth
rate of poverty as a result of migration is higher for HCR compared
to PG and SPG if the migrant’s income is very close (and below) the
poverty line.
2. We have obtained significant effects for both the policy variables, literacy rates and the indoor air pollution as significant with appropriate
signs (respectively negative and positive). However, the significance no
longer holds if we incorporate the endogenity of income growth rate
variable.
3. Except for the insignificance of the policy variables, both models with
and without endogeneity has more or less similar estimates of the coefficients. The estimates of GEP and IEP emphasizes many interesting
facts. It should be noted that since the elasticities depends on the initial inequality (gini) and ratio of poverty line and mean income, GEP
and IEP are not fixed. We have reported average values of GEP and
IEP over the rural and urban sectors of state regions, separately for
five time points, the period in which the data was collected. We have
noticed that the average GEP(IEP), lies mostly in the range of -1.5(0.5)
to -2.5(1.7), for HCR as the poverty index. GEP and IEP, are greater
when we consider the poverty index as PG and SPG. The higher values
of GEP and IEP for PG and SPG, has also been obtained in earlier
studies. This is due to the fact that these indices are more sensitive on
changes of income of the poor.
4. It has been observed that absolute values of average GEP(IEP) is
higher(lower) for the urban(rural) region. The positivity of IEP clearly
27
indicates that the force of poverty reduction is largely embedded due
to inequality. The reduction is greater in the urban region compared
to rural. The net differences between absolute values of GEP and IEP
is greater in the rural region.
5. In the less developed states, it has been observed that absolute values
of both GEP and IEP are lower. Alternatively, the force of poverty
reduction is largely embedded due to unequal incomes. The micro level
data sets used in this study, also allows us to evaluate the GEP and IEP
fluctuations within the state regions. Within states wide difference in
GEP and IEP has also been observed in Maharastha, Uttar Pradesh,
Madhya Pradesh, Odisha e,t,c has been widely noticed. In fact in
Odisha for Kalahanadi region we have obtained that GEP is very low
and IEP is negative. This regions faced a drought in the 1900s and still
now the income growth has been historically lower compared to others,
leading to this result.
28
References
Adams, R. (2004), “Economic growth, inequality and poverty: Estimating
the growth elasticity of poverty.” World Development, 32, 1989–2014.
Anselin, L. (2009), Spatial Econometrics: Methods and Models. Kluwer Academic Publishers, Dordrecht, Netherlands.
Bourguignon, F. (2003), “The growth elasticity of poverty reduction: explaining heterogeneity across countries and time periods.” In Inequality
and growth: theory and policy implications. (T. Eicher and S. Turnovsky,
eds.), MIT Press, Cambridge.
Chambers, D. and S. Dhongde (2011), “A non-parametric measure of poverty
elasticity.” Review of Income and Wealth, 57, 683–703.
Datt, G. and M. Ravallion (1992), “Growth and redistribution components of
changes in poverty measures: A decomposition with applications to brazil
and india in the 1980s.” Journal of Development Economics, 38, 275 –
295.
Deaton, A. (2003), “Adjusted indian poverty estimates for 1999-2000.” Economic and Political Weekly, 56, 322–326.
Driscoll, J. C. and A C. Kraay (1998), “Consistent covariance matrix estimation with spatially dependent panel data.” Review of Economics and
Statistics, 80, 549–560.
Du, Y., A. Park, and S. Wang (2005), “Migration and rural poverty in china.”
Journal of Comparative Economics, 33, 688–709.
Foster, J. E., J. Greer, and E. Thorbecke (1984), “A class of decomposable
poverty measures.” Econometrica, 52, 761–766.
Fosu, A.K. (2009), “Inequality and the impact of growth on poverty: Comparative evidence for sub-saharan africa.” The Journal of Development
Studies, 45, 726–745.
Gong, X., A. van Soest, and P. Zhang (2005), “The effects of the gender of
children on expenditure patterns in rural china: a semiparametric analysis.” Journal of Applied Econometrics, 20, 509–527.
Gundlach, E., de P.J. Navarro, and N. Weisert (2004), “Education is good for
the poor: A note on dollar and kraay.” In Inequality and growth: theory and
29
policy implications. (Shorrocks. A and R. van der Hoeven, eds.), Oxford
University Press, Oxford.
Hazan, M. and B. Berdugo (2002), “Child labour, fertility, and economic
growth.” The Economic Journal, 112, 810–828.
Hirway, I. (2003), “Identification of bpl households for poverty alleviation
programmes.” Economic and Political Weekly, 38, 4803–4038.
Kalwij, A. and A. Verschoor (2007), “Not by growth alone: The role of the
distribution of income in regional diversity in poverty reduction.” European
Economic Review, 51, 805–829.
Kelejian, H.H. and I.R. Prucha (1999), “A generalized moments estimator for
the autoregressive parameter in a spatial model.” International Economic
Review, 40, 509–533.
Kelejian, H.H. and I.R. Prucha (2004), “Estimation of simultaneous systems
of spatially interrelated cross sectional equations.” Journal of Econometrics, 118, 27–50.
Pradhan, J. (1993), “Drought in kalahandi-the real story.” Economic and
Political Weekly, 28.
Ram, R. (2007), “Roles of income and equality in poverty reduction: Recent
cross-country evidence.” Journal of International Development, 19, 919–
926.
Ravallion, M. (1995), “Growth and poverty: Evidence for developing countries in the 1980s.” Economics Letters, 48, 411–417.
Ravallion, M. (2011), “A comparative perspective on poverty reduction in
brazil, china, and india.” World Bank Economic Review, 26, 71–104.
Ravallion, M. and S. Chen (1997), “What can new survey data tell us about
recent changes in distribution and poverty?” World Bank Economic Review, 11, 357–382.
Ravallion, M. and G. Datt (1998), “Why have some indian states done better
than others at reducing rural poverty?” Economica, 65, 17–38.
Ravallion, M. and G. Datt (2002), “Why has economic growth been more
pro-poor in some states of india than others?” Journal of Development
Economics, 68, 381–400.
30
Sen, A.K. (1976), “Poverty: An ordinal approach to measurement.” Econometrica, 44, 219–231.
Shorrocks, A.F. (2013), “Decomposition procedures for distributional analysis: a unified framework based on the shapley value.” The Journal of
Economic Inequality, 11, 99–126.
Smith, K.R. (1993), “Fuel combustion, air pollution exposure, and health :
The situation in developing countries.” Annual Review of Energy and the
Environment, 18, 529–566.
Sung, Y.J. and S. Khagram (2005), “A comparative study of inequality and
corruption.” American Sociological review, 70, 136–157.
Zaman, K. and B.A. Khilji (2013), “The relationship between growthinequality-poverty triangle and pro-poor growth policies in pakistan: The
twin disappointments.” Economic Modelling, 30, 375–393.
31
9
Appendix
In a society, let at time point t, yt = (y1 , y2 , ..yq ...yn ) ∈ Rn++ , be the income
distribution, arranged in ascending order. where n is the number of individuals and q is the number of poor. The class of FGT index (Foster et al.,
1984) for the income distribution yt , with poverty line say z may be written
as follows
n
P
(z − yi )α I(yi ≤ z)
(10)
F GTα (yt ) = i=1
nz
For choice of α = 0, 1, 2, the poverty indices are HCR, PG and SPG
respectively. Let hcrt , pgt and spgt denotes the growth rate of the indices.
Let xt denotes the mean income of the poor.
Suppose at time point t+1, r number of individuals migrates from a different region. Also suppose r = r1 + r2 , r1 denotes the number of poor and
r2 denotes the number of non poor As a result of migration, let the mean
income of the poor changes to xt+1 . The following proposition relates the
growth rate of HCR and PG.
Proposition : hcrt ≥ pgt if and only if x̄t ≥ x̄t+1 .
Proof : It can be shown that P Gt = HCRt (1 − x̄t /z). Hence, the growth
rate may also be related as
1 + pgt = (1 + hcrt )θ
(11)
t+1
where θ = z−x̄
z−x̄t
Clearly the necessary and sufficient condition can be proved ⇐⇒ θ ≤ 1.
The equality holds only when θ = 1 or xt = xt+1 . Thus if all the r
migrants are non poor then also the equality holds.
32
Table 1: Descriptive Statistics : Rural India
statenames
Andhra Pardesh
Andhra Pardesh
Assam
Assam
Bihar
Bihar
Gujrat
Haryana
Haryana
Himachal Pradesh
Jammu & Kashmir
Karnataka
Karnataka
Karnataka
Karnataka
Kerala
Kerala
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Maharastra
Maharastra
Maharastra
Maharastra
Maharastra
Maharastra
Manipur
Manipur
Meghalaya
Orissa
Orissa
Orissa
Punjab
Punjab
Rajasthan
Rajasthan
Rajasthan
Rajasthan
Sikkim
Tamil Nadu
Tamil Nadu
Tamil Nadu
Tamil Nadu
Tripura
Uttar Pradesh
Uttar Pradesh
Uttar Pradesh
Uttar Pradesh
West Bengal
West Bengal
West Bengal
West Bengal
Arunachal Pradesh
Chandigarh
Delhi
Goa
Mizoram
Pondicheri
Chattisgarh
Uttarakhand
Jharkhand
state regions
coastal
Inland
plains east and west
Hills
Northen
Central
Eastern+Plains Northen
Eastern
Western
Himachal
Mountains
Coastal ANd Ghat
Inland Eastern
Inland Southern
Inland Northen
Northen
Southern
Vindhya
Central
Malwa
South Central
South Western
Northen
Coastal
Inland Western
Inland Northen
Inland Central
Inland Eastern
Eastern
Plains
Hills
Meghalaya
Coastal
Southern
Northen
Northen
Southern
Western
North Eastern
Southern
South Eastern
Sikkim
Coastal Northen
Coastal
Southern
Inland
Tripura
Western
Central
Eastern
Southern
Himalayan
Eastern plains
Central Plains
Western Plains
Arunachal Pradesh
Chandigarh
Delhi
Goa
Mizoram
Pondicheri
Chhattisgarh
Uttaranchal
Jharkhand
HCR
32.31
44.86
49.01
52.33
59.25
63.14
39.20
28.25
29.13
26.64
19.40
19.64
25.37
33.85
55.45
32.80
18.58
56.00
60.95
37.36
64.04
66.05
34.73
38.03
33.96
55.95
54.62
57.17
69.05
49.90
65.46
27.25
49.05
78.15
61.35
20.29
25.67
35.23
31.70
51.95
37.43
39.41
48.07
27.74
38.56
37.64
35.74
36.14
50.66
54.03
52.81
42.70
50.99
33.84
43.76
41.62
15.21
10.96
24.50
30.14
16.37
61.20
30.61
59.42
PGR
7.62
10.18
10.37
12.09
15.15
16.16
9.10
6.14
6.70
5.05
3.53
3.99
4.51
7.55
13.53
7.45
3.86
14.23
16.01
8.93
18.85
19.22
6.69
9.24
7.10
15.82
16.54
16.03
19.67
9.22
14.78
4.19
11.29
28.23
17.45
3.14
4.84
6.99
6.31
12.23
8.04
7.65
12.59
5.35
8.74
8.13
7.22
7.30
13.19
13.10
14.73
8.80
11.34
6.65
10.07
10.61
2.79
1.72
5.09
5.22
3.05
16.13
5.27
15.40
gini
27.21
25.94
22.22
19.27
21.38
20.85
24.73
29.97
28.35
28.41
22.28
26.14
21.62
24.42
22.20
30.21
35.18
23.49
24.61
28.26
29.79
23.95
23.07
29.24
25.92
27.47
28.51
26.34
24.86
15.68
17.58
19.75
23.65
23.23
25.81
28.27
27.29
22.49
22.25
25.93
22.95
24.30
30.18
25.70
25.00
30.38
21.40
25.99
25.20
25.06
30.06
20.81
23.98
23.62
25.85
30.34
24.88
25.32
27.71
20.03
30.30
24.89
27.30
22.75
MPCE
983.33
851.52
798.05
732.28
658.10
636.40
906.06
1299.12
1181.13
1117.34
1085.67
1030.64
882.72
839.42
660.35
1141.38
1619.07
669.91
654.34
856.60
669.11
604.27
830.83
1016.81
1029.30
827.02
820.20
806.40
720.21
921.78
835.23
898.12
653.96
462.53
586.07
1395.62
1299.00
963.85
984.41
875.64
935.92
949.09
812.04
924.83
815.82
920.75
827.76
891.73
747.90
738.15
793.53
766.94
736.40
839.78
796.29
1066.05
1619.58
1469.08
1524.34
1087.75
1231.87
649.87
1027.59
653.29
Electricity
69.34
80.94
34.72
26.72
5.75
18.38
81.24
87.46
84.78
95.81
93.81
73.20
79.54
81.25
80.08
68.99
79.70
54.69
66.78
78.76
64.97
78.95
58.22
81.60
79.69
72.27
76.18
71.10
59.60
86.93
64.68
53.67
42.76
16.74
24.93
94.14
93.35
46.27
59.56
41.80
66.76
91.21
81.84
73.23
79.32
79.55
58.06
32.95
11.59
25.74
24.40
24.56
24.92
37.30
25.95
52.44
89.43
96.98
98.05
70.00
78.23
59.34
64.38
22.95
F.Literacy
8.64
7.45
10.80
11.07
5.88
7.10
9.82
16.11
13.11
23.38
18.85
21.82
13.95
10.18
8.02
22.35
33.84
6.93
2.93
2.97
4.62
4.27
4.34
12.10
15.66
11.61
8.19
13.88
10.55
31.10
18.11
9.72
10.81
2.62
7.01
24.52
14.01
3.02
4.38
4.25
4.02
15.75
15.32
12.94
13.13
11.15
7.40
8.03
6.75
8.53
4.38
7.23
4.79
7.71
6.84
13.18
21.85
33.14
29.32
11.92
17.79
7.37
14.06
5.28
1
The Table contains average values of all the indicators
2
F.literacy implies Female literacy, C.Fuels implies percentage of households using non combustible cooking fuels.
33
cultivation
8.15
16.51
27.26
53.58
47.53
35.76
0.98
34.86
19.13
1.25
0.39
1.76
4.78
8.10
21.73
7.41
7.96
13.20
10.04
20.29
10.66
12.35
15.57
2.70
8.20
4.75
16.80
11.73
3.83
55.98
12.45
2.93
24.85
8.20
10.89
25.35
30.77
20.30
24.34
3.17
8.18
5.15
6.08
8.35
6.50
3.64
3.60
60.58
22.15
60.67
15.40
3.29
9.05
8.18
5.42
1.60
21.49
0.22
1.33
2.17
5.46
1.45
0.64
1.41
C.Fuels
83.49
88.94
91.81
92.83
87.33
93.73
83.87
82.30
84.74
80.85
84.52
82.32
91.59
86.47
95.14
88.08
80.86
98.02
97.21
92.57
96.76
94.45
98.19
77.82
74.92
70.75
65.69
88.16
90.48
69.51
93.70
97.11
89.57
96.80
94.14
69.19
75.33
95.33
94.81
95.04
94.56
67.28
80.40
89.59
86.78
76.84
95.31
94.46
96.53
90.79
99.06
96.17
81.40
82.92
80.27
88.42
18.14
11.34
45.44
82.30
74.04
97.37
80.14
96.84
Table 2: Descriptive Statistics : Urban India
statenames
Andhra Pardesh
Andhra Pardesh
Assam
Assam
Bihar
Bihar
Gujrat
Haryana
Haryana
Himachal Pradesh
Jammu & Kashmir
Karnataka
Karnataka
Karnataka
Karnataka
Kerala
Kerala
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Madhya Pradesh
Maharastra
Maharastra
Maharastra
Maharastra
Maharastra
Maharastra
Manipur
Manipur
Meghalaya
Orissa
Orissa
Orissa
Punjab
Punjab
Rajasthan
Rajasthan
Rajasthan
Rajasthan
Sikkim
Tamil Nadu
Tamil Nadu
Tamil Nadu
Tamil Nadu
Tripura
Uttar Pradesh
Uttar Pradesh
Uttar Pradesh
Uttar Pradesh
West Bengal
West Bengal
West Bengal
West Bengal
Arunachal Pradesh
Chandigarh
Delhi
Goa
Mizoram
Pondicheri
Chattisgarh
Uttarakhand
Jharkhand
state regions
coastal
Inland
plains east and west
Hills
Northen
Central
Eastern+Plains Northen
Eastern
Western
Himachal
Mountains
Coastal ANd Ghat
Inland Eastern
Inland Southern
Inland Northen
Northen
Southern
Vindhya
Central
Malwa
South Central
South Western
Northen
Coastal
Inland Western
Inland Northen
Inland Central
Inland Eastern
Eastern
Plains
Hills
Meghalaya
Coastal
Southern
Northen
Northen
Southern
Western
North Eastern
Southern
South Eastern
Sikkim
Coastal Northen
Coastal
Southern
Inland
Tripura
Western
Central
Eastern
Southern
Himalayan
Eastern plains
Central Plains
Western Plains
Arunachal Pradesh
Chandigarh
Delhi
Goa
Mizoram
Pondicheri
Chhattisgarh
Uttaranchal
Jharkhand
HCR
28.31
27.85
28.71
31.80
50.63
44.38
27.26
22.66
26.69
15.62
9.29
25.80
24.89
13.24
49.79
30.30
14.29
35.57
36.28
23.54
36.89
39.58
35.09
9.07
28.45
45.62
55.18
46.28
37.13
47.36
64.25
25.50
35.81
39.33
31.15
22.28
26.03
25.07
28.79
19.58
29.23
22.72
20.65
23.92
30.82
22.89
19.39
33.99
35.21
41.70
51.77
33.47
39.12
22.90
33.50
25.42
10.75
15.55
15.37
9.68
13.82
30.36
22.21
34.25
34
PGR
6.27
6.11
6.12
6.98
13.72
11.11
5.74
4.97
6.22
3.03
1.51
5.53
4.85
2.63
13.58
6.72
3.00
8.58
9.05
5.30
9.09
10.00
8.56
1.49
6.43
13.05
16.74
13.48
9.48
9.27
15.92
4.24
8.20
12.09
7.43
4.04
5.49
4.90
6.07
3.88
6.72
4.33
4.90
4.99
7.08
4.28
3.45
8.05
8.85
10.22
14.30
8.13
9.91
4.80
8.12
5.95
2.16
3.04
2.83
1.50
3.06
7.18
5.00
8.67
gini
36.01
34.00
29.93
31.57
30.42
31.02
30.44
32.11
30.01
37.74
27.19
35.12
25.95
31.07
30.18
37.23
39.46
31.43
38.18
34.52
35.05
30.70
30.72
34.13
37.41
33.41
33.86
36.13
28.34
18.91
15.84
24.47
34.69
35.24
31.09
31.96
32.87
27.92
35.35
28.24
30.87
24.04
36.38
31.04
34.87
34.54
29.77
34.77
35.91
30.94
28.53
30.56
32.67
36.52
33.39
28.39
40.77
36.34
33.10
22.00
30.87
32.66
30.60
34.51
MPCE
1709.70
1650.88
1438.55
1462.08
950.79
1053.80
1564.49
1806.76
1568.07
2137.30
1659.49
1726.58
1390.09
1964.11
1122.13
1487.54
2141.32
1157.88
1310.27
1512.02
1225.89
1076.97
1134.02
2315.84
1832.41
1336.43
1168.85
1374.77
1357.22
1081.56
911.65
1503.81
1216.41
1102.30
1173.68
1784.67
1714.22
1336.65
1523.98
1556.72
1369.06
1628.45
1742.69
1401.62
1349.42
1510.18
1428.41
1301.57
1359.13
1105.41
1006.81
1270.82
1235.13
1681.03
1372.66
1489.13
3172.57
2467.55
2090.59
1645.24
1615.69
1301.20
1518.17
1337.56
Electricity
88.19
94.00
84.54
79.32
46.05
76.67
93.84
94.47
93.59
96.35
99.06
95.20
91.77
94.62
88.58
86.50
90.41
91.10
95.88
97.42
93.48
95.46
93.14
97.82
93.64
92.56
92.30
93.03
89.99
94.38
96.01
96.17
78.06
68.19
78.46
97.72
97.41
90.84
91.54
94.35
94.92
97.07
93.00
88.50
91.91
90.98
88.32
81.97
78.63
78.61
70.92
80.04
71.46
84.64
73.17
92.94
95.28
98.37
96.90
96.21
91.72
89.47
95.24
81.06
F.Literacy
27.00
32.31
41.52
39.40
24.14
31.14
35.52
39.58
36.41
56.36
47.89
43.35
34.32
44.63
31.35
32.56
43.73
28.04
35.91
33.25
31.96
30.96
27.60
43.96
39.05
33.11
24.87
38.00
35.03
45.09
22.68
46.50
30.48
26.11
29.74
45.93
42.69
21.09
29.05
30.39
26.30
34.95
39.50
33.62
30.57
29.43
30.23
30.25
38.79
28.73
28.69
31.58
26.75
34.68
27.73
37.53
55.07
49.17
42.70
34.20
37.64
37.33
43.71
34.28
cultivation
0.72
1.27
0.61
0.96
3.70
4.10
0.10
4.24
2.48
0.04
0.04
0.14
1.20
0.90
4.59
1.71
2.18
1.75
1.64
1.70
1.43
1.87
3.23
0.37
0.92
0.75
2.21
2.40
0.42
37.52
0.19
0.02
0.98
0.81
0.73
2.85
3.65
1.00
1.74
0.17
0.58
0.06
1.00
1.12
1.40
0.81
0.15
5.17
1.57
2.96
1.13
0.05
0.28
0.59
0.19
0.06
13.70
0.30
0.73
1.04
3.61
0.14
0.04
0.06
C.Fuels
35.76
30.01
27.61
37.59
61.75
52.45
20.43
25.75
34.05
15.36
15.46
35.78
33.56
12.99
52.00
70.31
52.15
54.39
37.73
33.25
48.91
41.82
53.78
1.65
14.01
19.56
36.09
32.01
31.83
39.78
66.04
33.39
51.29
59.22
56.46
16.35
26.06
40.31
42.95
29.85
35.58
4.02
19.26
38.03
40.38
30.79
51.23
46.79
35.72
46.50
55.09
49.52
47.60
36.83
45.04
29.63
4.90
2.65
10.11
23.19
28.52
48.14
19.88
63.48
Table 3: Morans Test
Round 50
I
∆log(HCR) 0.07
∆log(P G)
0.06
∆log(M P CE)-0.04
∆log(gini) -0.12
electricity
0.55
education
0.32
cultivation -0.01
cooking
0.02
∆log(HCR) 0.08
∆log(P G)
0.06
∆log(M P CE)0.05
∆log(gini) -0.02
electricity
0.37
education
0.20
cultivation
0.14
cooking
0.22
∆log(HCR) 0.07
∆log(P G)
0.04
∆log(M P CE)0.01
∆log(gini) -0.04
electricity
0.23
education
0.00
cultivation
0.01
cooking
0.05
Round 55
Round 61
I
Pval
I
Rural India
0.14
0.28
0.00
0.26
0.18
0.27
0.00
0.30
0.40
0.13
0.04
0.19
0.11
0.14
0.03
0.14
0.00
0.60
0.00
0.54
0.00
0.40
0.00
0.50
0.47
-0.04
0.37
0.00
0.30
0.05
0.13
0.03
Urban India
0.11
0.05
0.22
0.03
0.17
0.02
0.31
0.03
0.22
-0.01
0.48
-0.08
0.46
0.01
0.36
0.01
0.00
0.51
0.00
0.40
0.00
0.07
0.14
0.19
0.03
0.13
0.03
0.19
0.00
0.19
0.00
0.08
Rural and Urban India
0.02
0.14
0.00
0.14
0.11
0.14
0.00
0.16
0.34
0.02
0.25
0.04
0.22
0.10
0.00
0.12
0.00
0.28
0.00
0.24
0.45
0.01
0.30
0.04
0.28
0.00
0.39
0.01
0.05
0.04
0.09
0.03
Pval
35
Pval
Round 66
I
Pval
0.00
0.00
0.01
0.02
0.00
0.00
0.43
0.23
0.10
0.07
0.02
0.09
0.56
0.37
-0.07
0.07
0.01
0.07
0.31
0.09
0.00
0.00
0.22
0.11
0.29
0.27
0.21
0.39
0.00
0.01
0.01
0.10
0.19
0.20
0.03
-0.01
0.34
-0.04
0.05
0.07
0.00
0.00
0.28
0.46
0.00
0.40
0.22
0.13
0.00
0.00
0.12
0.00
0.00
0.10
0.34
0.18
0.14
0.14
0.07
0.05
0.30
0.03
0.01
0.00
0.00
0.00
0.02
0.08
0.00
0.17
0.32
0.43
Table 4: Spatial Model
variables
y
y × i0
y × z/Y
g
g × i0
g × z/Y
f.literacy
cooking
ρ
1
2
3
−6.40a
7.22a
3.20a
4.64a
−2.65b
−3.78a
−0.05a
0.07c
HCR
PG
SPG
(0.30) −8.31a (0.27) −9.82a (0.31)
(1.10) 10.53a (1.01) 12.72a (1.18)
(0.17)
3.01a (0.15)
3.10a (0.16)
a
(0.16)
5.32 (0.40)
5.81a (0.65)
(1.30)
-2.44 (1.72)
-1.81 (2.48)
a
(0.18) −3.80 (0.11) −3.94a (0.10)
(0.01) −0.05a (0.02)
-0.04 (0.03)
a
(0.04)
0.09 (0.02)
0.10b (0.05)
0.17a (0.03)
0.11a (0.03)
0.10a (0.02)
Notes : Estimated results based on a Spatial Autoregressive
model with Driscoll Karry Standard errors. ζ ∗ is the residual collected from the first stage regression.
Set of Instruments are electricity consumption, average cultivated
lands, % of households using non combustible cooking materials,
Female literacy rates (Secondary Level) and MPCE from NSSO
employment Unemployment rounds.
The notations for the first six variables are similar to equation 1. In the parenthesis we report standard errors. a, b and
c implies significance at 1%, 5%, 10% respectively.(Two tailed test)
36
Table 5: Endogenity Tests
variables
y
y × i0
y × z/Y
g
g × i0
g × z/Y
f.literacy
cooking
ζ∗
ρ
1
2
3
-1.29
-5.56
1.18
3.06a
1.09
−3.03a
−0.06a
0.08c
−5.21a
HCR
PG
(1.94)
2.42 (4.88)
(4.90) -16.32 (12.32)
(0.79)
-1.24 (1.93)
(0.48)
2.01 (1.61)
(1.63)
5.41 (4.08)
(0.30) −2.23a (0.76)
(0.01) −0.07a (0.03)
(0.04)
0.11a (0.02)
(1.75) −10.93b (4.89)
0.17a (0.04)
0.11a (0.03)
SPG
1.55 (7.21)
-15.74 (18.18)
-1.41 (2.87)
2.30 (2.43)
6.52 (6.17)
−2.28b (1.14)
-0.07 (0.04)
0.12a (0.04)
-11.59 (7.58)
0.09a (0.02)
Notes : Estimated results based on a Spatial Autoregressive model
with Driscoll Karry Standard errors. ζ ∗ is the residual collected from
the first stage regression.
Set of Instruments are electricity consumption, average cultivated
lands, % of households using non combustible cooking materials,
Female literacy rates (Secondary Level) and MPCE from NSSO employment Unemployment rounds.
The notations for the first six variables are similar to equation 1.
In the parenthesis we report standard errors. a, b and c implies
significance at 1%, 5%, 10% respectively.(Two tailed test)
37
Table 6: Spatial Model with endogenous income growth rate
variables
y
y × i0
y × z/Y
g
g × i0
g × z/Y
f.literacy
cooking
ρ
1
2
3
−6.41a
7.30a
3.20a
4.62a
-2.67
−3.77a
-0.05
0.07
HCR
PG
SPG
(0.86) −8.31a (1.04) −9.83a (1.36)
(2.31) 10.55a (2.80) 12.74a (3.64)
(0.49)
3.01a (0.59)
3.10a (0.77)
a
(0.66)
5.32 (0.80)
5.81a (1.03)
(1.72)
-2.45 (2.07)
-1.81 (2.70)
a
(0.40) −3.79 (0.48) −3.94a (0.63)
(0.04)
-0.05 (0.05)
-0.04 (0.06)
c
(0.05)
0.09 (0.05)
0.10 (0.07)
0.20b (0.08)
0.12c (0.07)
0.10 (0.07)
Notes : Estimated results based on a Spatial Autoregressive model
with Driscoll Karry Standard errors. ζ ∗ is the residual collected from
the first stage regression.
Set of Instruments are electricity consumption, average cultivated
lands, % of households using non combustible cooking materials, Female literacy rates (Secondary Level) and MPCE from NSSO employment Unemployment rounds.
The notations for the first six variables are similar to equation 1.
In the parenthesis we report standard errors. a, b and c implies
significance at 1%, 5%, 10% respectively.(Two tailed test)
38
Table 7: Predicted GEP and IEP for Rural and Urban India
Year
Rural India
1993-94
1999-00
2004-05
2009-10
Urban India
1993-94
1999-00
2004-05
2009-10
1
2
3
HCR
GEP
PG
SPG
HCR
IEP
PG
SPG
-1.52(0.52)
-1.70(0.55)
-1.88(0.49)
-2.08(0.48)
-2.69(0.56)
-2.87(0.56)
-3.04(0.54)
-3.22(0.52)
-3.52(0.63)
-3.70(0.62)
-3.87(0.61)
-4.07(0.58)
0.51(0.59)
0.72(0.66)
0.93(0.54)
1.16(0.55)
1.24(0.60)
1.46(0.67)
1.67(0.55)
1.91(0.55)
1.78(0.63)
2.00(0.71)
2.22(0.57)
2.47(0.58)
-1.88(0.48)
-2.19(0.38)
-2.25(0.41)
-2.34(0.47)
-2.91(0.51)
-3.20(0.46)
-3.25(0.48)
-3.33(0.49)
-3.68(0.57)
-3.98(0.53)
-4.03(0.55)
-4.12(0.55)
1.14(0.55)
1.50(0.39)
1.57(0.44)
1.67(0.57)
1.89(0.56)
2.25(0.39)
2.33(0.44)
2.43(0.58)
2.48(0.59)
2.85(0.41)
2.93(0.46)
3.03(0.61)
Notes : GEP and IEP are predicted from equation 3 and 4. The value of the parameters
are from the spatial model with additional endogenous variable.
Set of Instruments are electricity consumption, average cultivated lands, % of households
using non combustible cooking materials, Female literacy rates (Secondary Level) and MPCE
from NSSO employment Unemployment rounds.
The notations for the first six variables are similar to equation 1. In the parenthesis we report
standard errors. a, b and c implies significance at 1%, 5%, 10% respectively.(Two tailed test)
39
Arunachal Pradesh
Assam
Bihar
Chandigarh
Chhattisgarh
Delhi
Goa
Gujrat
Haryana
Himachal Pradesh
Jammu & Kashmir
Jharkhand
Karnataka
Kerala
Madhya Pradesh
Maharastra
Manipur
Meghalaya
Mizoram
Orissa
Pondicheri
Punjab
Rajasthan
Sikkim
Tamil Nadu
Tripura
Uttar Pradesh
Uttaranchal
West Bengal
-1 0
1
2
3
-1 0
1
2
3
-1 0
1
2
3
-1 0
1
2
3
-1 0
1
2
3
Andhra Pardesh
rs
gep for hcr
iep for hcr
Graphs by States of India
Figure 1: In the diagram we have reported the estimated values of Growth
and Inequality elasticity of poverty, see equation 3 and 4, for HCR. The
values of αi & βi , ∀i ∈ 1, 2, 3, are obtained from Table 6 Further for all the
regions, GEP and IEP are based on their initial inequality and average values
of the ratio of mean income and poverty line. For each state, bar diagrams
on the left and the right are respectively for rural and urban state regions.
Absolute values of the GEP are presented in the Figure. The black portion
of the bars indicate the difference between GEP and IEP.
40